Method for detecting metal content of gold and iron ores

Abstract

The application belongs to the technical field of nondestructive testing, and discloses a gold ore and iron ore metal content detection method; the method comprises the following steps: XRF energy spectrum is acquired by using an X-ray tube, LIBS spectrum is collected by using a double-pulse Nd:YAG laser, and Raman spectrum is acquired by using a semiconductor laser; baseline correction is performed on the LIBS spectrum, Savitzky-Golay smoothing pretreatment is implemented on the Raman spectrum, and three types of spectral features are extracted respectively; a physical constraint network model is constructed, the corrected content of Fe / Au elements, the fusion weight and the LIBS dominant factor are outputted after the XRF, LIBS and Raman features are inputted; finally, the content of the gold ore and the iron ore is calculated by combining the XRF, LIBS fusion weight and network output parameters through a dynamic weight fusion strategy; the application realizes a technical breakthrough of multi-element collaborative detection and complex matrix interference suppression, and effectively improves the detection precision of Fe and Au.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and more specifically, to a method for detecting the metal content of gold and iron ores. Background Technology

[0002] In the process of mineral resource exploration and beneficiation, rapid and accurate detection of the content of key elements such as Fe and Au in ore is crucial for resource assessment and process optimization.

[0003] Chinese patent application CN111272686A discloses a hyperspectral method for detecting the iron grade of iron ore beneficiation powder, comprising the following steps: S1. Establishing a hyperspectral reference database of iron ore beneficiation powder with different iron grade levels; S2. Determining the strong linearity identification band of the hyperspectral curves of iron ore beneficiation powder with different iron grades; S3. Establishing a hyperspectral prediction model of the spectral reflectance of the strong linearity identification band of the hyperspectral curve and the iron grade of the beneficiation powder; S4. Determining the iron grade of the sample to be tested. This invention solves the technical problems of traditional methods for determining the iron ore powder grade, such as cumbersome workflow, long cycle, chemical reagent contamination, or poor applicability.

[0004] While the above methods can meet the needs of most scenarios, research and practical application of these methods and existing technologies have revealed at least the following shortcomings:

[0005] Traditional single-spectral techniques are limited by matrix interference, detection limits, and multi-element synergistic analysis capabilities, making it difficult to meet the online detection requirements of complex ore components. When complex coexisting elements or matrix effects exist in the ore, their anti-interference ability is insufficient and their adaptability to complex matrices is weak, which can easily affect the accuracy of the detection results.

[0006] In view of this, the present invention proposes a method for detecting the metal content of gold and iron ores to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a method for detecting the metal content of gold and iron ore, comprising the following steps:

[0008] XRF energy spectrum was acquired using an X-ray tube, LIBS spectrum was acquired using a dual-pulse Nd:YAG laser, and Raman spectrum was acquired using a semiconductor laser.

[0009] Baseline correction was performed on the LIBS spectrum to obtain the corrected LIBS spectrum; Savitzky-Golay smoothing was performed on the Raman spectrum to obtain the corrected Raman spectrum;

[0010] Feature extraction was performed on the collected XRF energy spectrum, modified LIBS spectrum and modified Raman spectrum to obtain XRF features, LIBS features and Raman features;

[0011] Using XRF features, LIBS features, and Raman features as inputs to the physical constraint network, an output set is obtained. The output set includes the corrected Fe content corresponding to the XRF features, the corrected Fe content corresponding to the LIBS features, the fusion weights corresponding to the XRF features, the fusion weights corresponding to the LIBS features, the corrected Au content corresponding to the XRF features, the corrected Au content corresponding to the LIBS features, and the LIBS dominant factor.

[0012] Based on the fusion weights corresponding to XRF features and LIBS features, and using a dynamic weight fusion strategy combined with the output set of the physical constraint network, the gold and iron ore contents are calculated.

[0013] Furthermore, training methods for physically constrained networks include:

[0014] K sets of training data are collected in advance. The training data includes input data and output set. The input data includes XRF features, LIBS features and Raman features.

[0015] The input data is used as the input to the physical constraint network (PCLN), and the output set is used as the output of the PCLN. The goal is to minimize the error between the output set and the actual output set. A natural heuristic optimization algorithm is used to optimize the PCLN parameters to obtain the network parameters that minimize the error between the output set and the actual output set. The PCLN constructed using these parameters is then trained. The PCLN includes a physical constraint layer for applying hard and soft constraints to Fe and Au elements. The hard constraint incorporates the Fe matrix interference model and the Saha-Boltzmann equation, and achieves hard constraint on Fe through residual calculation. The hard constraint also applies to Au based on nanoparticle etching kinetics. The soft constraint applies to Fe through the correction relationship between crystallization pressure and the modified mass absorption coefficient. Furthermore, the soft constraint uses a Sigmoid function to process the signal-to-noise ratio of the Au LIBS signal intensity, obtaining dynamic weights to dynamically adjust the Au LIBS signal intensity within the PCLN.

[0016] Furthermore, methods for obtaining Raman features include:

[0017] The full width at half maximum (FWHM) of the characteristic peak intensity corresponding to Fe in the corrected Raman spectrum is obtained, and the crystallization pressure is calculated based on the FWHM of the characteristic peak intensity corresponding to Fe; the crystallization pressure is then used as a Raman characteristic.

[0018] Furthermore, methods for obtaining XRF features include:

[0019] The characteristic peak intensities corresponding to Fe and Au elements in the XRF energy spectrum were obtained respectively. The background intensity was subtracted from the characteristic peak intensities corresponding to Au elements in the XRF energy spectrum signal to obtain the metallic intensity of Au elements in the XRF energy spectrum.

[0020] A predetermined number of standard samples containing different Fe concentrations and predetermined interfering elements were prepared. XRF spectra were acquired on the standard samples. Based on the acquired XRF spectra, the characteristic peak intensities of Fe were extracted. These intensities were then combined with the Fe concentration, the concentration of the predetermined interfering element, and the excitation energy of Fe. The mass absorption coefficient is used to establish the Fe matrix interference model;

[0021] Correction of Fe element in excitation energy based on crystallization pressure inversion. The mass absorption coefficient below;

[0022] AuNPs substrates were prepared, and the Raman signal intensity of probe molecules on the AuNPs substrates was measured. The number of analyte molecules adsorbed on the AuNPs surface was calculated by fluorescence labeling. The Raman signal intensity of probe molecules of the same concentration on ordinary glass slides was measured as the Raman signal intensity of pure analytes, and the AuNPs enhancement factor was calculated.

[0023] The XRF characteristics are obtained by splicing together the characteristic peak intensity of Fe, the metallic intensity of Au, the corrected mass absorption coefficient, and the AuNPs enhancement factor in the XRF spectrum.

[0024] Furthermore, methods for obtaining LIBS features include:

[0025] The characteristic peak intensities corresponding to the primary and secondary ionization states of Fe in the modified LIBS spectrum were obtained respectively. The plasma temperature was calculated based on the characteristic peak intensities corresponding to the primary and secondary ionization states of Fe and the Saha-Boltzmann equation.

[0026] Obtain the characteristic peak intensity corresponding to the Au element, and perform background dynamic suppression by using the preset inner caliber intensity to obtain the normalized characteristic peak intensity corresponding to the Au element.

[0027] The plasma temperature and the normalized characteristic peak intensities corresponding to Au elements are spliced ​​together to form the LIBS feature.

[0028] Furthermore, methods for obtaining iron ore content include:

[0029] The iron ore content is obtained by weighting the Fe content corresponding to the XRF feature, the Fe content corresponding to the LIBS feature, the fusion weight corresponding to the XRF feature, and the fusion weight corresponding to the LIBS feature. When the signal-to-noise ratio corresponding to the XRF feature is not less than the preset first signal-to-noise ratio threshold, the fusion weight corresponding to the XRF feature is forced to be equal to the preset first weight threshold.

[0030] Furthermore, methods for determining gold ore content include:

[0031] The gold content is obtained by weighting the corrected Au content corresponding to the XRF feature, the corrected Au content corresponding to the LIBS feature, and the LIBS dominant factor. The sum of the weights of the corrected Au content corresponding to the XRF feature and the corrected Au content corresponding to the LIBS feature is 1, and the fusion weight corresponding to the XRF feature is the value of the LIBS dominant factor. When the signal-to-noise ratio corresponding to the XRF feature is less than the preset second signal-to-noise ratio threshold, the LIBS dominant factor is forced to be equal to the preset second weight threshold. When the gold content is greater than the preset concentration threshold, the LIBS dominant factor is forced to gradually change from the second weight threshold to the third weight threshold.

[0032] Furthermore, the acquisition timing of the XRF spectrum and the LIBS spectrum is controlled to meet timing control constraints; the Raman spectrum is acquired during the interval between the XRF spectrum and the LIBS spectrum; the method for controlling the acquisition timing of the XRF spectrum and the LIBS spectrum to meet timing control constraints includes:

[0033] The acquisition timing of XRF and LIBS spectra satisfies the following: after the initial excitation of the LIBS spectrum, the duration of the confinement interval and plasma lifetime is delayed before the excitation and acquisition of the XRF spectrum are triggered.

[0034] Furthermore, methods for obtaining modified LIBS spectra include:

[0035] A convex optimization objective function is constructed based on a non-parametric model of convex optimization to perform baseline adaptive correction of the LIBS spectrum, thereby obtaining the corrected LIBS spectrum.

[0036] Furthermore, methods for obtaining corrected Raman spectra include:

[0037] For each data point in the Raman spectrum, take a region with a width of [missing information] centered on that data point. The window is used to calculate the smoothing value of the data points through polynomial fitting. The smoothing value of all data points in the Raman spectrum is calculated by iterating through the window to obtain the corrected Raman spectrum.

[0038] The technical effects and advantages of the method for detecting metal content in gold and iron ores according to the present invention are as follows:

[0039] This invention utilizes pulse timing control to achieve synchronous multispectral acquisition, meeting the rapid detection needs of mineral processing sites. The method for detecting metal content in gold and iron ores based on spectral analysis integrates multimodal data from XRF, LIBS, and Raman spectroscopy, combined with a physical constraint network and dynamic weight fusion strategy, achieving a technological breakthrough in multi-element collaborative detection and suppression of interference from complex matrices. By embedding hard and soft constraints into the physical constraint network, absorption-enhancement interference is effectively suppressed. Furthermore, the dynamic weight fusion strategy allocates the contributions of XRF and LIBS in real time according to spectral quality, effectively improving the detection accuracy of Fe and Au. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of a method for detecting the metal content of gold and iron ore according to the present invention.

[0041] Figure 2 This is a schematic diagram of the time-series acquisition of XRF energy spectrum, LIBS spectrum and Raman spectrum of the present invention;

[0042] Figure 3 This is a schematic diagram of the method for obtaining XRF features according to the present invention;

[0043] Figure 4 This is a schematic diagram of a method for enhancing physical constraint networks according to the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1

[0046] Please see Figure 1 As shown in this embodiment, a method for detecting the metal content of gold and iron ore includes the following steps:

[0047] XRF energy spectrum was acquired using an X-ray tube, LIBS spectrum was acquired using a dual-pulse Nd:YAG laser, and Raman spectrum was acquired using a semiconductor laser; among which, reference... Figure 2The acquisition timing of XRF and LIBS spectra is controlled to meet timing constraints; Raman spectroscopy is acquired during the intervals between XRF and LIBS spectroscopy. Using an X-ray tube, a dual-pulse Nd:YAG laser, and a semiconductor laser to acquire XRF, LIBS, and Raman spectra respectively, and with reasonable control of the acquisition timing, plays a crucial role in the detection of metal content in gold and iron ores. XRF spectroscopy provides information on the type and content of elements, LIBS spectroscopy is beneficial for trace element analysis and valence state determination, and Raman spectroscopy reflects mineral crystal structure and chemical bond information. The three spectroscopic techniques complement each other, enabling comprehensive acquisition of ore characteristics. Controlling the acquisition timing avoids interference from plasma generated by LIBS excitation in XRF spectroscopy acquisition, ensuring that XRF spectroscopy is acquired under stable conditions. Acquiring Raman spectroscopy during the intervals avoids interference with other spectral signals, ensuring the accuracy and purity of the three spectral data. This lays a solid foundation for subsequent feature extraction, interference model construction, and accurate calculation of element content, improving the accuracy and reliability of metal content detection in gold and iron ores.

[0048] Methods for controlling the acquisition timing of XRF and LIBS spectra to meet timing control constraints include:

[0049] The acquisition timing of XRF and LIBS spectra satisfies the following: after the initial excitation of the LIBS spectrum, the confinement interval and plasma lifetime are delayed before triggering the excitation and acquisition of the XRF spectrum; for example, the XRF spectrum acquisition time... ;in, The acquisition time for the LIBS spectrum; The constraint interval; This refers to the plasma lifetime.

[0050] By controlling the acquisition timing of XRF and LIBS spectra, and delaying the XRF acquisition by the duration of "constraint interval + plasma lifetime" after the initial LIBS excitation, interference from the plasma generated by LIBS excitation on XRF acquisition can be avoided. This ensures that XRF acquisition is performed after the plasma has decayed to a stable state, improving the purity of both types of spectral data, reducing signal-noise superposition, and making the extraction of characteristic signals of elements such as Fe and Au more accurate. This provides a high-quality data foundation for subsequent processing such as matrix interference model construction and physical constraint network analysis, ultimately significantly improving the accuracy and reliability of gold and iron ore metal content detection.

[0051] Baseline correction is performed on LIBS spectra to obtain corrected LIBS spectra; Savitzky-Golay smoothing is performed on Raman spectra to obtain corrected Raman spectra. Baseline correction of LIBS spectra adjusts slowly changing baselines to zero baselines, eliminating data errors caused by baseline shift, drift, and background interference, reducing the impact of noise on spectral signals, and making elemental characteristic peaks in LIBS spectra more prominent and accurate, thus improving the accuracy of elemental characteristic identification and content detection. Savitzky-Golay smoothing of Raman spectra reduces random fluctuations in spectral data through polynomial fitting, preserving important features such as spectral shape, peaks, and curvature as much as possible while reducing noise, resulting in smoother and cleaner Raman spectra. This facilitates more accurate analysis of mineral crystal structures, chemical bonds, and other information, providing a more reliable data foundation for the detection of metal content in gold and iron ores, and improving the accuracy and reliability of detection results.

[0052] Methods for obtaining modified LIBS spectra include:

[0053] A nonparametric model based on convex optimization is used to construct a convex optimization objective function to perform baseline adaptive correction on the LIBS spectrum, resulting in the corrected LIBS spectrum; the convex optimization objective function is as follows: ;in, Based on baseline To optimize the object, solve for the objective function that makes it convex. Minimized value; The original signal; As the baseline; It is a second-order difference matrix; For regularization parameters, Describing the L2 norm, This represents the L1 norm.

[0054] Baseline adaptive correction of LIBS spectra using a non-parametric model based on convex optimization can effectively eliminate interference such as baseline drift and noise in the original signal. This makes the characteristic peaks of elements such as Fe and Au in the corrected LIBS spectra clearer and more prominent, avoiding misjudgment of characteristic peaks or deviation in intensity calculation caused by baseline interference. This provides pure and reliable data for subsequent feature extraction, plasma parameter inversion, and element content calculation, thereby improving the accuracy of gold and iron ore metal content detection and ensuring that the analysis results of metal content in ores more accurately reflect the true composition.

[0055] Methods for obtaining corrected Raman spectra include:

[0056] For each data point in the Raman spectrum, take a region with a width of [missing information] centered on that data point. The window is used to calculate the smoothing value of the data points through polynomial fitting. This process is repeated for all data points in the Raman spectrum to obtain the corrected Raman spectrum. The polynomial fitting formula is as follows: ;in, The intensity of the smoothed Raman spectrum; These are the original data points within the window; These are the fitting coefficients; Normalization factor; It is half the width of the window.

[0057] Savitzky-Golay smoothing was applied to the Raman spectra. By taking a window with a width of 2m+1 around each data point and performing polynomial fitting, random noise in the spectrum was effectively reduced while preserving the peak position, intensity, and shape characteristics. This significantly improved the signal-to-noise ratio of the Raman spectra, making the chemical state information of Au in gold ore and the mineral structure information of Fe in iron ore more clearly discernible. This provides reliable data support for subsequent Raman-based mineral phase analysis, chemical bond identification, and pressure correction, thereby improving the inversion accuracy of the XRF / LIBS fusion model for Fe and Au elemental content and ensuring the accuracy and stability of metal content detection in complex ore matrices.

[0058] Feature extraction was performed on the acquired XRF, modified LIBS, and modified Raman spectra to obtain XRF, LIBS, and Raman features, respectively. This process allows for the extraction of ore composition and structural information from multiple dimensions: XRF features provide direct evidence for elemental concentration inversion, LIBS features reveal elemental ionization and plasma states, and Raman features reflect mineral crystal structure and chemical environment. After normalization and dimensionality reduction, these features are used as input to a physical constraint network, effectively integrating multispectral information, suppressing the limitations of single technologies, and providing a comprehensive and complementary feature space for accurate calculation of Fe and Au content. This significantly improves the accuracy and reliability of metal content detection in complex ore matrices.

[0059] Methods for obtaining Raman features include:

[0060] The full width at half maximum (FWHM) of the characteristic peak intensity corresponding to Fe in the corrected Raman spectrum is obtained, and the crystallization pressure is calculated based on the FWHM of the characteristic peak intensity corresponding to Fe. The crystallization pressure is then used as a Raman characteristic, such as crystallization pressure. ;in, Crystallization pressure; is the full width at half maximum (FWHM) of the intensity of the characteristic peak of pyrite in the Raman spectrum.

[0061] Reference Figure 3Methods for obtaining XRF features include:

[0062] The characteristic peak intensities corresponding to Fe and Au elements in the XRF energy spectrum were obtained respectively. The background intensity was subtracted from the characteristic peak intensities corresponding to Au elements in the XRF energy spectrum signal to obtain the metallic intensity of Au elements in the XRF energy spectrum.

[0063] A predetermined number of standard samples containing different Fe concentrations and predetermined interfering elements were prepared. XRF spectra were acquired on the standard samples. Based on the acquired XRF spectra, the characteristic peak intensities of Fe were extracted. These intensities were then combined with the Fe concentration, the concentration of the predetermined interfering element, and the excitation energy of Fe. The mass absorption coefficient is used to establish a Fe matrix interference model, such as... ;in, The characteristic peak intensity of Fe element extracted from the XPF energy spectrum of the standard sample; These are instrument constants and can be obtained empirically. This represents the concentration of Fe element; For Fe element in excitation energy The mass absorption coefficient below; Fe and interfering elements in the excitation energy The mass absorption coefficient below; For elements Interference factors; Interference element The concentration;

[0064] Correction of Fe element in excitation energy based on crystallization pressure inversion. The mass absorption coefficient below; such as ;in, For Fe element in excitation energy The corrected mass absorption coefficient is as follows;

[0065] AuNPs substrates were prepared, and the Raman signal intensity of probe molecules on the AuNPs substrates was measured. The number of analyte molecules adsorbed on the AuNPs surface was calculated using a fluorescence labeling method. The Raman signal intensity of probe molecules at the same concentration on a regular glass slide was measured as the Raman signal intensity of the pure analyte, and the AuNPs enhancement factor was calculated. (The AuNPs enhancement factor is then described.) ;in, The Raman signal intensity on the AuNPs substrate; The number of analyte molecules adsorbed on the surface of AuNPs; The Raman signal intensity of the pure analyte; This represents the number of analyte molecules in the bulk solution.

[0066] The XRF characteristics are obtained by splicing together the characteristic peak intensity of Fe, the metallic intensity of Au, the corrected mass absorption coefficient, and the AuNPs enhancement factor in the XRF spectrum.

[0067] Multidimensional feature extraction was performed on XRF spectra. The net intensity of Au was obtained by subtracting background intensity. A Fe matrix interference model was established based on standard samples, and the Fe mass absorption coefficient was corrected by crystallization pressure. The surface enhancement effect was quantified by AuNPs enhancement factors. Finally, an XRF feature vector containing Fe / Au characteristic peak intensities, matrix interference parameters, pressure correction coefficients, and enhancement factors was formed. This achieved multi-physical quantity synergistic constraint: the matrix interference model suppressed the absorption / enhancement interference of complex coexisting elements on the Fe signal, the crystallization pressure correction improved the explanatory power of mineral structure on the XRF absorption coefficient, and the AuNPs enhancement factor enhanced the detection sensitivity of trace Au. This provided high-precision, multi-dimensional XRF feature input for subsequent physical constraint networks, significantly improving the accuracy and anti-interference ability of gold and iron ore metal content detection.

[0068] Methods for obtaining LIBS features include:

[0069] The characteristic peak intensities corresponding to the primary and secondary ionization states of Fe in the modified LIBS spectrum were obtained respectively. Based on the characteristic peak intensities corresponding to the primary and secondary ionization states of Fe, the plasma temperature was calculated using the Saha-Boltzmann equation. ;like ;in, The characteristic peak intensity corresponding to the primary ionization state of Fe element; The characteristic peak intensity corresponding to the secondary ionization state of Fe element; Boltzmann's constant; This represents the excitation energy of the primary ionization state of Fe. The excitation energy for the secondary ionization state of Fe element; The statistical weights for the primary ionization states of Fe; The statistical weights for the secondary ionization states of Fe; The transition probability of the primary ionization state of Fe element; The transition probability of the second ionization state of Fe element; The wavelength of the spectral line of the primary ionization state of Fe; The wavelength of the spectral line of the secondary ionization state of Fe element;

[0070] The characteristic peak intensity corresponding to Au is obtained, and background dynamic suppression is performed by using a preset internal standard intensity to obtain the normalized characteristic peak intensity corresponding to Au; for example, the preset internal standard intensity is the internal standard intensity of Mn at 405.5nm. Calculate the normalized characteristic peak intensity corresponding to Au element. ; The LIBS signal strength of Au element;

[0071] The plasma temperature and the normalized characteristic peak intensities corresponding to Au elements are spliced ​​together to form the LIBS feature.

[0072] Feature extraction was performed on the modified LIBS spectrum, and the plasma temperature was calculated using the Saha-Boltzmann equation. The intensity of the Au characteristic peak was normalized using the Mn internal standard, forming a LIBS feature vector that incorporates both plasma state and trace element signals. This approach achieves dual optimization: plasma temperature correction eliminates differences in ionization state distribution caused by laser energy fluctuations, ensuring the accuracy of Fe concentration inversion; internal standard normalization dynamically suppresses matrix background interference, significantly improving the signal-to-noise ratio of Au detection. These two methods work together to provide reliable plasma parameters and trace element characteristics for the physical constraint network, effectively overcoming the technical bottleneck of LIBS signals being susceptible to temperature fluctuations and background noise in complex ore matrices, ultimately improving the accuracy and stability of gold and iron ore metal content detection.

[0073] XRF, LIBS, and Raman features are used as inputs to a physical constraint network to obtain an output set. The output set includes the corrected Fe content corresponding to the XRF features, the corrected Fe content corresponding to the LIBS features, the fusion weights corresponding to the XRF features, the fusion weights corresponding to the LIBS features, the corrected Au content corresponding to the XRF features, the corrected Au content corresponding to the LIBS features, and the LIBS dominant factor. By inputting XRF, LIBS, and Raman features into the physical constraint network and embedding the XRF fundamental parameter equations, the Saha-Boltzmann equation, and the mineral structure model, physical consistency constraints on multimodal data are achieved. The network output set uses the fusion of XRF and LIBS Fe / Au content predictions combined with dynamic weights to allocate the optimal data contribution in real time, while simultaneously outputting the LIBS dominant factor to optimize the plasma state model. This approach overcomes the limitations of single-spectral techniques: physical constraints ensure that elemental content inversion conforms to physical laws such as X-ray absorption and plasma ionization; dynamic weight fusion enhances the complementarity of multi-source data under complex matrices; and the LIBS dominant factor strengthens the detection reliability of trace Au, providing intelligent and interference-resistant core algorithm support for the detection of metal content in gold and iron ores.

[0074] Training methods for physically constrained networks include:

[0075] K sets of training data are collected in advance. The training data includes input data and output set. The input data includes XRF features, LIBS features and Raman features.

[0076] The input data is used as the input to the physical constraint network, and the output set is used as the output of the physical constraint network. The goal is to minimize the error between the output set and the actual output set. A natural heuristic optimization algorithm is used to optimize the network parameters of the physical constraint network to obtain the network parameters that minimize the error between the output set and the actual output set. The physical constraint network constructed using these parameters is then used as the trained physical constraint network. The physical constraint network includes a physical constraint layer, which is used to apply hard and soft constraints to the Fe and Au elements.

[0077] The hard constraint is embedded in the Fe matrix disturbance model and the Saha-Boltzmann equation, and the hard constraint on the Fe element is achieved through residual calculation, such as the hard constraint of the Fe element. ;in, and These are all weight parameters, which can be obtained through natural heuristic optimization algorithms; The residuals of the Fe matrix interference model; The hard constraint is based on the plasma temperature residual; the hard constraint is also based on the nanoparticle etching dynamics to hard constrain the Au element, such as... ;in, For parameters related to surface plasmon resonance; and They are respectively and ion concentration; The reaction rate constant is defined as follows: The above hard constraints can force the network output to conform to the physical laws of X-ray absorption, plasma ionization and surface plasma resonance, effectively suppressing absorption-enhancement interference and plasma fluctuation errors under complex substrates.

[0078] The soft constraint applies to Fe element through a correction relationship between crystallization pressure and the modified mass absorption coefficient. Furthermore, the soft constraint uses a Sigmoid function to process the signal-to-noise ratio of Au element LIBS signal intensity, obtaining dynamic weights to dynamically adjust the Au element LIBS signal intensity in the physical constraint network, such as dynamic weighting. ;in, The signal-to-noise ratio of the LIBS signal intensity of Au element; soft constraints can achieve flexible adjustment of the detection process by mineral structure characteristics and spectral quality, and improve the reliability of Fe occurrence state identification and trace Au signal enhancement in ores of different geological origins.

[0079] Based on the fusion weights corresponding to XRF features and LIBS features, and using a dynamic weight fusion strategy combined with the output set of the physical constraint network, the gold and iron ore contents are calculated; this enables intelligent collaboration of multispectral data.

[0080] Methods for obtaining iron ore content include:

[0081] The iron ore content is obtained by weighting the corrected Fe content corresponding to XRF features, the corrected Fe content corresponding to LIBS features, the fusion weights corresponding to XRF features and LIBS features, and so on. ;in, The corrected content of Fe element corresponding to the XRF feature; These are the fusion weights corresponding to the XRF features; The corrected content of Fe element corresponding to the LIBS feature; The fusion weights corresponding to the LIBS features are defined as follows: When the signal-to-noise ratio corresponding to the XRF features is not less than the preset first signal-to-noise ratio threshold (e.g., 50), the fusion weights corresponding to the XRF features are forced to be equal to the preset first weight threshold (e.g., 0.8), which can ensure that high-reliability data dominates the calculation of Fe content and effectively suppress the interference of LIBS plasma fluctuations on Fe detection. When the XRF signal is weak, the real-time ionization analysis advantage of LIBS is brought into play by adaptively adjusting the weights. This mechanism can improve the accuracy of Fe content detection. At the same time, by complementing the Au metal intensity of XRF with the normalized signal of LIBS, the accuracy of Au content detection is improved, and the quantitative analysis capability of multi-metal elements in complex mineral phases is significantly enhanced.

[0082] Methods for determining gold ore content include:

[0083] Gold content is obtained by weighting the corrected Au content corresponding to XRF features, the corrected Au content corresponding to LIBS features, and the LIBS dominant factor. ;in, The corrected content of Au element corresponding to the XRF feature; It is the dominant factor in LIBS; The corrected content of Au corresponding to the LIBS feature; the sum of the weights of the corrected content of Au corresponding to the XRF feature and the corrected content of Au corresponding to the LIBS feature is 1, and the fusion weight corresponding to the XRF feature is the value of the LIBS dominant factor; when the signal-to-noise ratio corresponding to the XRF feature is less than the preset second signal-to-noise ratio threshold (e.g., 2), the LIBS dominant factor is forced to be equal to the preset second weight threshold (e.g., 1), which can avoid noise interference from XRF in trace Au detection; when the gold content is greater than the preset concentration threshold, the LIBS dominant factor is forced to gradually change from the second weight threshold (e.g., 1) to the third weight threshold (e.g., 0.6), which can balance the matrix interference of XRF and the ionization fluctuation risk of LIBS under high concentration; the above method can improve the detection accuracy and quantitative accuracy of Au in complex ore matrices, while effectively suppressing the absorption and shielding effect of high Fe matrix on Au signal, significantly improving the accuracy and reliability of Au content detection in gold resource assessment and beneficiation process.

[0084] Example 2

[0085] Reference Figure 4 This embodiment provides a method for enhancing physical constraint networks to constrain Au. 0 / Au 3+ Proportional prediction includes the following steps:

[0086] The characteristic peak intensities corresponding to Au elements in different chemical states were calculated based on the matrix interference model.

[0087] The spectral intensities of Au in different chemical states were calculated using the Saha-Boltzmann equation;

[0088] XRF spectra (energy-theoretical spectral intensity) and LIBS spectra (wavelength-theoretical spectral intensity) are classified by chemical state and stored in the database;

[0089] During the training of the physical constraint network, the difference between the predicted spectral intensity and the theoretical spectral intensity stored in the database is calculated. The loss function is constructed with the goal of minimizing the difference between the predicted spectral intensity and the theoretical spectral intensity stored in the database, and is added to the hard constraint of the Au element in the physical constraint network.

[0090] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0091] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting the metal content of gold and iron ores, characterized in that, Includes the following steps: XRF energy spectrum was acquired using an X-ray tube, LIBS spectrum was acquired using a dual-pulse Nd:YAG laser, and Raman spectrum was acquired using a semiconductor laser. Baseline correction was performed on the LIBS spectrum to obtain the corrected LIBS spectrum; Savitzky-Golay smoothing was performed on the Raman spectrum to obtain the corrected Raman spectrum; Feature extraction was performed on the collected XRF energy spectrum, modified LIBS spectrum, and modified Raman spectrum to obtain XRF features, LIBS features, and Raman features; Using XRF features, LIBS features, and Raman features as inputs to the physical constraint network, an output set is obtained. The output set includes the corrected Fe content corresponding to the XRF features, the corrected Fe content corresponding to the LIBS features, the fusion weights corresponding to the XRF features, the fusion weights corresponding to the LIBS features, the corrected Au content corresponding to the XRF features, the corrected Au content corresponding to the LIBS features, and the LIBS dominant factor. Training methods for physically constrained networks include: K sets of training data are collected in advance. The training data includes input data and output set. The input data includes XRF features, LIBS features and Raman features. The input data is used as the input to the physical constraint network, and the output set is used as the output of the physical constraint network. The goal is to minimize the error between the output set and the actual output set. A natural heuristic optimization algorithm is used to optimize the network parameters of the physical constraint network to obtain the network parameters that minimize the error between the output set and the actual output set. The physical constraint network constructed with these parameters is used as the trained physical constraint network. The physical constraint network includes a physical constraint layer for applying hard and soft constraints to Fe and Au elements. The hard constraint embeds the Fe matrix interference model and the Saha-Boltzmann equation, and achieves hard constraint on Fe through residual calculation. The hard constraint also applies to Au based on nanoparticle etching dynamics. The soft constraint applies to Fe through the correction relationship between crystallization pressure and the modified mass absorption coefficient. Furthermore, the soft constraint uses the Sigmoid function to process the signal-to-noise ratio of the Au LIBS signal intensity, obtaining dynamic weights to dynamically adjust the Au LIBS signal intensity in the physical constraint network. Based on the fusion weights corresponding to XRF features and LIBS features, and using a dynamic weight fusion strategy combined with the output set of the physical constraint network, the gold and iron ore contents are calculated.

2. The method for detecting the metal content of gold and iron ore according to claim 1, characterized in that, Methods for obtaining Raman features include: The full width at half maximum (FWHM) of the characteristic peak intensity corresponding to Fe in the corrected Raman spectrum is obtained, and the crystallization pressure is calculated based on the FWHM of the characteristic peak intensity corresponding to Fe. The crystallization pressure is then used as a Raman characteristic.

3. The method for detecting the metal content of gold and iron ore according to claim 1, characterized in that, Methods for obtaining XRF features include: The characteristic peak intensities corresponding to Fe and Au elements in the XRF energy spectrum were obtained respectively. The background intensity was subtracted from the characteristic peak intensities corresponding to Au elements in the XRF energy spectrum signal to obtain the metallic intensity of Au elements in the XRF energy spectrum. A predetermined number of standard samples containing different Fe concentrations and predetermined interfering elements were prepared. XRF spectra were acquired on the standard samples. Based on the acquired XRF spectra, the characteristic peak intensities of Fe were extracted. These intensities were then combined with the Fe concentration, the concentration of the predetermined interfering element, and the excitation energy of Fe. The mass absorption coefficient is used to establish the Fe matrix interference model; Correction of Fe element in excitation energy based on crystallization pressure inversion. The mass absorption coefficient below; AuNPs substrates were prepared, and the Raman signal intensity of probe molecules on the AuNPs substrates was measured. The number of analyte molecules adsorbed on the AuNPs surface was calculated by fluorescence labeling. The Raman signal intensity of probe molecules of the same concentration on ordinary glass slides was measured as the Raman signal intensity of pure analytes, and the AuNPs enhancement factor was calculated. The XRF characteristics are obtained by splicing together the characteristic peak intensity of Fe, the metallic intensity of Au, the corrected mass absorption coefficient, and the AuNPs enhancement factor in the XRF spectrum.

4. The method for detecting the metal content of gold and iron ore according to claim 1, characterized in that, Methods for obtaining LIBS features include: The characteristic peak intensities corresponding to the primary and secondary ionization states of Fe in the modified LIBS spectrum were obtained respectively. The plasma temperature was calculated based on the characteristic peak intensities corresponding to the primary and secondary ionization states of Fe and the Saha-Boltzmann equation. Obtain the characteristic peak intensity corresponding to the Au element, and perform background dynamic suppression by using the preset inner caliber intensity to obtain the normalized characteristic peak intensity corresponding to the Au element. The plasma temperature and the normalized characteristic peak intensities corresponding to Au elements are spliced ​​together to form the LIBS feature.

5. The method for detecting the metal content of gold and iron ore according to claim 1, characterized in that, Methods for obtaining iron ore content include: The iron ore content is obtained by weighting the Fe content corresponding to the XRF feature, the Fe content corresponding to the LIBS feature, the fusion weight corresponding to the XRF feature, and the fusion weight corresponding to the LIBS feature. When the signal-to-noise ratio corresponding to the XRF feature is not less than the preset first signal-to-noise ratio threshold, the fusion weight corresponding to the XRF feature is forced to be equal to the preset first weight threshold.

6. The method for detecting the metal content of gold and iron ores according to claim 1, characterized in that, Methods for determining gold ore content include: The gold content is obtained by weighting the corrected Au content corresponding to the XRF feature, the corrected Au content corresponding to the LIBS feature, and the LIBS dominant factor. The sum of the weights of the corrected Au content corresponding to the XRF feature and the corrected Au content corresponding to the LIBS feature is 1, and the fusion weight corresponding to the XRF feature is the value of the LIBS dominant factor. When the signal-to-noise ratio corresponding to the XRF feature is less than the preset second signal-to-noise ratio threshold, the LIBS dominant factor is forced to be equal to the preset second weight threshold. When the gold content is greater than the preset concentration threshold, the LIBS dominant factor is forced to gradually change from the second weight threshold to the third weight threshold.

7. The method for detecting the metal content of gold and iron ore according to claim 1, characterized in that, The method for controlling the acquisition timing of the XRF spectrum and the LIBS spectrum to meet timing control constraints includes: Raman spectroscopy is acquired during the intervals between the XRF spectrum and the LIBS spectrum; The acquisition timing of XRF and LIBS spectra satisfies the following: after the initial excitation of the LIBS spectrum, the duration of the confinement interval and plasma lifetime is delayed before the excitation and acquisition of the XRF spectrum are triggered.

8. The method for detecting the metal content of gold and iron ore according to claim 1, characterized in that, Methods for obtaining corrected LIBS spectra include: A convex optimization objective function is constructed based on a non-parametric model of convex optimization to perform baseline adaptive correction of the LIBS spectrum, thereby obtaining the corrected LIBS spectrum.

9. The method for detecting the metal content of gold and iron ore according to claim 1, characterized in that, Methods for obtaining corrected Raman spectra include: For each data point in the Raman spectrum, take a region with a width of [missing information] centered on that data point. A window is used to calculate the smoothing value of the data points within the window through polynomial fitting. This process is repeated for all data points in the Raman spectrum to obtain the corrected Raman spectrum. It is half the width of the window.

Images